JPH01234940A - Preference type queue management - Google Patents

Abstract

PURPOSE: To manage the priority queue efficiently by storing entry of the highest priority to a storage register. CONSTITUTION: Every time a new element is written in a queue, the priority of the new element is compared with the priority of an existing element in a storage register 10. When the new element has higher priority, the existing element in the storage register 10 is written at a head of a stack 16. On the other hand, when the element in the storage register 10 has higher priority, the new element is written at a head of the stack. Thus, it is warranted that the storage register 10 has an element with highest priority at all times. Thus, the priority queue is managed efficiently.

Description

DETAILED DESCRIPTION OF THE INVENTION A. Field of Industrial Application This invention relates generally to data processing architectures;
More specifically, it concerns the architecture of data queues.

B. Prior Art and its Problems In many data processing applications, data is associated with an inherent priority for its manipulation. For example, in local area network contention resolution, where multiple nodes connected to the network must compete for access to a common resource, the data transmission requirements are critical to the overall operation of the network. It makes sense to provide preferred services to nodes that are uniquely more important. Typically, the prior art stored descriptors specifying the number of sessions to be processed by a node in a prioritized queue. As each new session and associated priority is assigned to a node, the session must be organized in priority order relative to the priorities of existing sessions at that node. Traditionally, this is done by examining the contents of a prioritized queue representing multiple existing sessions and writing the new session's descriptor to the queue in a position ranked according to its priority. was. Therefore, after writing a new session's descriptor to the priority queue at a node,
The queue will be reorganized with the newly assigned priority.

Problems arise with this mode of operation because the occurrence of assignments for new sessions is random in time. Because new events or sessions to be recorded in a priority queue occur asynchronously, it is difficult for traditional priority queuing mechanisms to quickly and repeatedly respond to entries of new data into the queue. becomes. This is because each entry requires an assignment of the relative priority of the new element to the priority of existing elements in the priority queue.

There are other data processing applications where priority queuing is used. For example, multiple programming applications. One type of multiprogrammed computer architecture is one in which multiple programs are programmed onto a single resource CPU by imposing time-sliced operations on each program's execution.
The work step architecture is such that it is executed in

Time is divided into periods and each of the programs to be executed by the CPU is assigned a relative priority for execution. At the beginning of each time-sliced period, the priority of each program is examined and the program with the highest priority is selected for execution during that time-sliced period. , where programs or subroutines are assigned on an event-driven basis, the growth of priority queues representing programs to be executed is done in such a way that it is asynchronous to the time slicing operations of multiple programming processes. must be

A priority queue that tracks which program has the highest priority for execution in the next cycle is reorganized each time a new program or subroutine has an assigned priority for that queue. If it had to be done, fast repetition of the assignment of new programs to the priority queue would not be possible, and at the same time the correctly identified highest priority program would be available for execution in the next cycle. I can't even do that. It is therefore an object of the present invention to manage priority queues in a more efficient manner than previously.

Another object of the present invention is to manage priority queues faster than previously possible.

Another object of the invention is to enable faster asynchronous allocation of new elements to priority queues than previously possible.

Another object of the invention is to make the elements at the head of a priority queue readily available in an improved manner.

SUMMARY OF THE INVENTION To achieve this objective, the method of the present invention for managing priority queues in a data processing system includes: (a) processing data portions and queues in response to requests for highest priority data; (b) reading the first element pointed to by the stack top pointer among a plurality of elements in the stack, each having a priority section and a data section, and determining its contents; The step of writing to the above holding register and (
c) traversing the multiple elements in the stack by finding the first entry in the stack pointed to by the stack search pointer, which is initially set equal to the top of stack pointer; (d) comparing the contents of the entry with the contents of the holding register; (e) swapping the contents of an entry in the stack pointed to by the stack search pointer with the contents of the holding register; (e) determining whether new data is to be written to the stack; and (r) (g) comparing the new data to be written to the stack to determine whether the priority portion of the new data is higher than the priority portion of the contents of the holding register; and (g) the priority of the new data. If the priority part is higher than the priority part of the contents of the holding register, the contents of the stack top pointer are incremented and the contents of the holding register are transferred to the contents of the stack newly pointed to by the stack top pointer. writing the new data to the holding register;
h) writing the new data onto the stack if the priority portion of the new data is not higher than the priority portion of the contents of the holding register; and (i)
repeating the steps after step (c) until the comparison of all entries in the stack is completed, so that the entry with the highest priority is stored in the holding register. There is.

Hereinafter, the operation of the present invention will be explained along with examples.

D. Examples First, examples of the present invention will be outlined. According to the invention,
In processing a priority queue, the elements are stored in an unmemorized stack and are searched for the next highest priority element only after the highest priority element has been read from its holding register. be done. The present invention can do this.

This is because each time a new element is written to this queue, a comparison is made between the priority of the new element and the priority of the existing elements in the holding register. If the new element has a higher priority, the existing element in the holding register is written to the top of the stack. On the other hand, if the element in the holding register has a higher priority, the new element is written to the top of the stack. Therefore, it is guaranteed that the holding register will always contain the element with the highest priority. As a result, reading a priority queue by reading the contents of a holding register ensures that the highest priority element is there and can be accessed immediately without further searching the queue. As previously mentioned, search operations are only involved after reading the contents of the holding register. To provide fast read operations, the top element of the stack is first placed in a holding register. A priority search operation is then performed such that each remaining element of the priority queue has a priority that is compared to the priority of the contents of the holding register. If the priority of the element in the holding register is lower than the priority of the element in the priority queue,
A swap operation is always performed. By the time the search operation reaches the end of the stack, the holding register contains the queue element with the highest priority.

Hereinafter, embodiments will be described in detail with reference to the drawings.

The fast access priority queue according to the present invention operates in data processing applications where there is some regularity as to when the priority queue is read. This is illustratively illustrated in a timing frame such as that shown in FIG. 4 for a local area network such as that shown in FIG. In a local area network such as that shown in FIG. and node 3°) are logically interconnected. Each node includes a network interface unit 5 (hereinafter also referred to as "network controller") and a local device 7. The local device 7 is, for example, a computer, a data interface or other data processing and communication device.

In many LAN applications, each node has a network controller 5 for managing multiple communication sessions with multiple other nodes. Each communication session to be managed by the network interface unit 5 has an associated priority value assigned to it. This value indicates the inherent relative importance that a communication session has with respect to other communication sessions to be managed by that node 3 and by other nodes 3° in the system. The present invention is particularly useful in such applications.

In order to give every node and every session of the network an opportunity to transmit their information, time is divided into a series of time frames (see Figure 4). 1
Each time frame includes a control section with the same 1t7J header, an acknowledgment section, and a conflict resolution section. This frame has a data section following a control section. The data part comprises the relevant data to be transmitted via the communication medium 1 to a destination node in the system. Typically, the control portion of the frame is, for example, 40 microseconds, and the data portion of the frame is of variable duration, for example, from 40 microseconds to 400 microseconds. In order for a node to select possible sessions in the contention resolution portion of the time frame control portion of FIG. 4, network interface unit 5 must maintain a priority queue.

A priority queue maintains a stack as shown in FIG. 3 containing an entry having two parts. The first part of one entry in the stack of Figure 3 is the priority value for that entry, and during the conflict resolution part the other priority values for other sessions are the priorities to be compared. is the value of In addition to the priority portion of the entries in the stack of FIG. 3, there is a data portion associated with this priority portion. In the example described here, this data portion may be an address pointer. This address pointer is stored in association with the network interface unit 5 and points to the data to be transmitted on the selected session. The data portion of the entry into the stack 16 of FIG. 3 may be actual data. For example, it may be the actual data part of a message packet to be transmitted over the communication network 1. The invention is particularly useful in these types of applications.

Figure 2 shows the controller 9 of the fast access priority queue.
FIG. As can be seen from the figure, the queue controller 9 selects the session with the highest priority using the bus controller 15 and the data output to the communication network 1 when the node in question wins the conflict resolution. The memory management processor 11 and the session data RAM are interconnected to each other. Queue controller 9 maintains a stack 16 with multiple entries. Each entry includes the entry's priority and one address pointer.

If its queue and function select inputs indicate to the queue controller 9 that the highest priority element is to be read from the stack and output, then that highest priority element will be the priority element supplied to the bus controller 15. portion and a pointer portion provided to memory management processor 11 from the stack. Memory management processor 11 uses the pointer portion to identify the data segment in session data RAM 13 that should be transmitted over the communication link if the conflict resolution is won. The bus controller 15 uses the priority values to participate in bus contention resolution operations for the communication network 1.

The queue controller 9 shown in FIG. 2 has a queue state machine that implements the process shown in the flowchart of FIG.

The queue state machine includes a top of stack pointer 12, a stack search pointer 14, and a RAM control (this is stack 1
6). The queue pointer 9 further includes a holding register 10, a -hour register 18,
Includes input register 22 and comparison logic 20.

Next, the operation will be explained. Assume that the stack 16 of FIG. 3 stores a plurality of stack entries, each containing a priority portion and a data portion.

The tail of the stack has a fixed value. For example, the address value may be O. The top of the stack is stack top pointer 1
2 (hereinafter referred to as r I max J). In the time frame of FIG. 4, at the end of the acknowledgment portion of the control, for example, the queue controller 9 via the queue and function selection input lines:
The receiver takes one input requesting that the highest priority entry be accessed from stack 16. At this time,
In the flowchart of FIG. 1, the holding register 10 that currently contains the highest priority entry is read and its value is output from the controller's I10 driver to the bus controller 15 and memory management controller via the f'-9 bus. The signal is supplied to the processor 11. Then, as shown in the flowchart of FIG. 1, the contents of the stack at the top of the stack indicated by the stack top pointer 12 are read. This value is the value of the entry newly written to the holding register 10. At this time, the value of the stack top pointer 12 is decremented by one unit, indicating that the top of the stack is at the next lower address value.

The contents of holding register 10 are immediately available for access by queue controller 9 according to the invention.

The queue controller 9 was thus able to respond immediately to the queue and function selection requests due to the end of the acknowledgment portion of the time frame.
until the next control part of the next time frame occurs.
It has a waiting time. It is during this wait time that the stack 16 is searched to identify the highest priority element and placed in the holding register 10. This is accomplished by a single loop as shown in the flowchart of FIG. This loop begins by sequentially searching the stack 16 using the stack search pointer 14, which points to the top of the stack by having the same contents as the stack top pointer 12. The contents of our priority part are held in register 1
It is compared with the priority value of the entry stored as 0.

If the former is higher than the latter, a swap operation must be performed to swap the contents of the holding register 10 with the contents of the stack at the location pointed to by the stack search pointer 14. This is done by taking the contents of the entry pointed to by the stack search pointer 14 and placing it in a temporary register 18 of the queue controller 9. The contents of holding register 10 are then written to the stack at the location pointed to by stack search pointer 14. Then, the contents of the - hour register 18 are written into the holding register 1o. When this swap operation is performed, the contents of the holding register may have the highest priority entry for any entry in the stack. The loop in Figure 1 is completed,
By the time the comparison is completed for all entries in the stack, the contents of the holding registers ensure that all entries in the stack 16 have the highest priority.

As can be seen from the flowchart of FIG. The process proceeds to the step of checking whether a new entry to be written to the stack 16 has been entered into the queue controller 9.

Proceed to this step also if the swap operation described above is completed. As shown in Figure 1, input register 2
2 is determined as to whether any new data is ready to be added to the stack. input register 2
2 is determined to have a new entry containing both a priority part and a pointer part, a comparison operation must be performed as to whether this new entry has a higher priority than that contained in the holding register 10. Must be. Thus, comparison logic 20 compares the priority portion of the contents of input register 22 with the priority portion of the contents of holding register 10 and, if the former is higher than the latter, new data is written to holding register 10. It will be done.

This is done as follows. First, the value Imax of the stack top pointer 12 is incremented by one. Next, the contents of the holding register 10 are transferred to the stack top pointer 12.
is written to the location in the stack pointed to by . The new data in input register 22 is then written to holding register 10. In this way, there is a high probability that the holding register will contain an entry with the highest priority value of any value in the stack.

On the other hand, if the new data has a priority part that is not higher than the contents of the holding register 10, then this new data is written to the stack. This is done as follows. The value Imax of the stack top pointer 12 is incremented by one.

The new data in input register 22 is then written to the location in the stack pointed to by stack top pointer 12. The flowchart then returns to "B". Point "B" is the point reached if there were no new data inputs to the queue controller 9. At point "B" a determination is made as to whether this comparison process has been performed on the entire stack. The stack search pointer 14 is given a variable name ■. In this flowchart, it is checked whether the value of ■ is zero. Tore means to check whether the end of the stack has been reached. When we actually reach the tail of the stack, the flowchart is at point “A”
and waits until the new data is ready to be written to the queue controller 9 or until the next control part of the time frame occurs. On the other hand, when the value of I in the stack search pointer 14 is not zero, the value of
The loop continues by making a determination as to whether the next entry in stack 16 has a higher priority value than holding register 10. In this way, upon completion of the loop, it is guaranteed that the contents of holding register 10 will have a higher priority value than the priority value of any entry in stack 16. The contents of holding register 10 are readily available to applications such as the LAN priority conflict resolution application U described above.

FIG. 6 shows another application of the invention in a data processor based on a working step architecture. In a work step architecture, multiple programming operations occur where each of multiple programs or program segments is given a priority value and its time is divided into a series of equal time intervals. Before each time interval, the program or program segment with the highest priority is selected for execution during the next interval. An exemplary computer architecture for performing the operations of the work step architecture is shown in FIG. The CPU is a shared data processing resource and is connected to a queue control dispatcher. The queue control dispatcher has a timer input that divides its time into a series of equal time intervals. A priority stack, such as stack 16 described above, is connected to the queue control dispatcher. A large amount of memory is also connected to the queue control dispatcher. This memory has address inputs for that connection and the program or program segment identified by the queue control dispatcher as having the highest priority (so that this program is selected for execution in the next time interval). be remembered). The queue hunter 9 of FIG. 2 is included in the queue control dispatcher of FIG.

The queue controller 9 has the highest priority and has a queue function selection input for requesting identification of the next program to be selected for execution in the next time interval. It operates in the same manner as described for the LAN application. The operation of the queue controller 9 in the application of the work step architecture of FIG. 6 is much the same as that described for the LAN application of FIG. In particular, if the assignment of a new program or program segment to the queue control dispatcher of FIG. 6 is asynchronous with the occurrence of a series of time intervals set by the timer of FIG. can be adapted to. This is because the queue controller 9 makes available in the holding register 10 the identification of the highest priority program available at the beginning of each time interval, and the queue controller 9 also makes available in the priority stack 16 the identity of the highest priority program available at the beginning of each time interval. This is because it is always available for adding new entries.

DETAILED DESCRIPTION OF THE F3 INVENTION According to the present invention, priority queues can be managed more efficiently than in the past because the highest priority entry is stored in a holding register.

[Brief explanation of the drawing]

FIG. 1 is a detailed flow diagram of the present invention, and FIG.
FIG. 3 is a diagram showing an example of a stack and holding register according to the present invention; FIG. 4 is a diagram showing a typical data communication time diagram to which the present invention can be applied. A frame diagram, FIG.
FIG. 6 is a diagram illustrating the architecture of a work chip data processor capable of managing priority assignments in accordance with the present invention. Applicant International Business Machines Corporation Representative Patent Attorney Takashi Tonmiya - (1 other person) 2% l1ai-1]2 Troller 0 Motion lT 1 Talking Shisukuzu2kkua1-Less Lee Aikone I Branch Tie-Tai 14 Aoshi・Nsu 7 Sword Divination l°Sc, ヮ7th 3rd
Figure bowl (#nateri communication time frame calculation 4 Mouth roll η Lueriane 1.I゛ Worker business Steuber N Techrama and use 3 jig 10 ・7
+ru46 figure

Claims (1)

Claims: A method for managing a priority queue in a data processing system, comprising: (a) reading the contents of a holding register having a data portion and a priority portion in response to a request for highest priority data; (b) reading the first element pointed to by the stack head pointer among a plurality of elements in the stack, each having a priority portion and a data portion, and writing its contents to the holding register; Search multiple elements in the stack by finding the first entry in the stack pointed to by the stack search pointer, which is initially set equal to the stack top pointer, and search for the entry pointed to by the stack search pointer. (d) comparing the contents of said holding register with the contents of said stack search pointer if the priority portion of the contents of said entry in said stack is higher than the priority portion of said contents of said holding register; (e) determining whether new data is to be written to the stack; (g) comparing new data to be written to the stack to determine whether the priority portion is higher than the priority portion of the contents of the holding register; incrementing the contents of the stack top pointer and writing the contents of the holding register to a location in the stack newly pointed to by the stack top pointer if the content of the holding register is higher than the priority part; further writing the new data to the holding register; (h) writing the new data to the stack if the priority portion of the new data is not higher than the priority portion of the contents of the holding register; (i) repeating the steps from step (c) onward until the comparison of all entries in the stack is completed, so that the entry with the highest priority is stored in the holding register; A priority queue management method characterized by: